Author name cluster

Yuchen Lin

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

7 papers

1 author row

EAAI Journal 2026 Journal Article

Dynamic patient similarity modeling with multi-source fused clinical knowledge for enhanced disease prediction

Yichen He
Yuchen Lin
Xiaorou Zheng
Shoubin Dong
Jun Fu

Accurate clinical disease prediction is crucial for modern healthcare. However, current methods for clinical disease prediction using multi-source Electronic Health Records (EHRs) are limited by coarse-grained integration and static utilization of similar patient data, failing to capture inter-entity interactions or dynamic disease evolution. In terms of the contribution to artificial intelligence, this paper proposes MFaDP, a novel framework integrating multi-source clinical knowledge including external knowledge graphs, medical code hierarchies, and internal co-occurrence patterns to construct multi-dimensional knowledge subgraphs, and pre-train high-quality entity representations via self-supervised reconstruction tasks for patient modeling and disease prediction. Crucially, MFaDP pioneers the dynamic modeling of similar patient evolution, leveraging historical visit records with the clinical pathways of similar patients, with an attention mechanism to dynamically extract reference information and uncover latent pathological patterns. Regarding the application in engineering, the proposed framework is applied to the critical task of intelligent clinical decision support. Extensive experiments on two large-scale public datasets, MIMIC-III and MIMIC-IV, demonstrate that MFaDP significantly outperforms state-of-the-art models across multiple prediction tasks, validating its advanced capability in harnessing complex EHR data for enhanced predictive performance.

Details DOI

AIIM Journal 2026 Journal Article

IKDP: Implicit Knowledge Enhanced Disease Prediction via heterogeneous admission sequence graphs

Zongbao Yang
Yuchen Lin
Yichen He
Jinlong Hu
Ruxin Wang
Hao Zhang
Shoubin Dong

Despite significant advances in deep learning for electronic health record (EHR) modeling, accurately representing complex disease relationships and admission trajectories remains challenging. Current approaches that leverage external knowledge graphs to learn patient representations are often limited by incomplete knowledge coverage. Furthermore, these methods frequently overlook implicit information within patient data, such as inter-patient similarities and latent disease correlations, and often discard patients with only a single admission, thereby losing valuable clinical insights. To address these limitations, we introduce the Implicit Knowledge Enhanced Disease Prediction model (IKDP) via heterogeneous admission sequence graphs (SeqGs), which harnesses implicit knowledge from comprehensive patient admission data. IKDP integrates an auxiliary pre-training strategy with end-to-end optimization to effectively process multi-dimensional patient data and compute inter-patient similarities as complementary knowledge. Specifically, the model constructs SeqGs for each patient, which capture complex disease dependencies and the dynamic evolution of health status. Moreover, critical paths extracted from the SeqGs, combined with similar patient analysis and historical admission records, are utilized to elucidate the reasoning behind predictions. The code is available at https: //github. com/SCUT-CCNL/IKDP.

Details DOI

NeurIPS Conference 2025 Conference Paper

PartCrafter: Structured 3D Mesh Generation via Compositional Latent Diffusion Transformers

Yuchen Lin
Chenguo Lin
Panwang Pan
Honglei Yan
Feng Yiqiang
Yadong Mu
Katerina Fragkiadaki

We introduce PartCrafter, the first structured 3D generative model that jointly synthesizes multiple semantically meaningful and geometrically distinct 3D meshes from a single RGB image. Unlike existing methods that either produce monolithic 3D shapes or follow two-stage pipelines, i. e. first segmenting an image and then reconstructing each segment, PartCrafter adopts a unified, compositional generation architecture that does not rely on pre-segmented inputs. Conditioned on a single image, it simultaneously denoises multiple 3D parts, enabling end-to-end part-aware generation of both individual objects and complex multi-object scenes. PartCrafter builds upon a pretrained 3D mesh diffusion transformer (DiT) trained on whole objects, inheriting the pretrained weights, encoder, and decoder, and introduces two key innovations: (1) A compositional latent space, where each 3D part is represented by a set of disentangled latent tokens; (2) A hierarchical attention mechanism that enables structured information flow both within individual parts and across all parts, ensuring global coherence while preserving part-level detail during generation. To support part-level supervision, we curate a new dataset by mining part-level annotations from large-scale 3D object datasets. Experiments show that PartCrafter outperforms existing approaches in generating decomposable 3D meshes, including parts that are not directly visible in input images, demonstrating the strength of part-aware generative priors for 3D understanding and synthesis. Code and training data are released.

PDF Details

JBHI Journal 2025 Journal Article

Pretraining-based Relevance-aware Visit Similarity Network for Drug Recommendation

Yichen He
Shoubin Dong
Yuchen Lin
Xiaorou Zheng
Jinlong Hu

Drug recommendation based on electronic health records (EHR) relies heavily on precise patient modeling, which is more complex than conventional recommendation tasks as it requires both temporal modeling of disease progression and referencing similar patients' medication information. However, sparse visit records and vague patient similarity in EHR data pose significant challenges, often introducing noise and ambiguity. To address the above challenges, we propose RaVSNet ( R elevance a ware V isit S imilarity Net work), which improves drug recommendation by leveraging both longitudinal and transversal visit similarity and integrating medical relevance knowledge. RaVSNet utilizes multi-dimensional visit information similar to the patient's current visit as a reference, and employs a relevance-aware network to explicitly model the matching relationships between medical conditions and medications. Additionally, RaVSNet designs a general pretraining framework specifically for drug recommendation, including two tasks, Medication Sequence Reconstruction (MSR) and Causal Effect Inference (CEI), to discover the deep connections between medical information and medications. Experimental results on two public EHR datasets, MIMIC-III and MIMIC-IV demonstrate that the proposed algorithm outperforms state-of-the-art methods, yielding more accurate drug recommendation combinations, and the proposed general pretraining framework can be seamlessly integrated into most drug recommendation methods to achieve performance improvements. The implementation is available at: https://github.com/SCUT-CCNL/RaVSNet.

Details DOI

YNIMG Journal 2024 Journal Article

Different oscillatory mechanisms of dementia-related diseases with cognitive impairment in closed-eye state

Talifu Zikereya
Yuchen Lin
Zhizhen Zhang
Ignacio Taguas
Kaixuan Shi
Chuanliang Han

The escalating global trend of aging has intensified the focus on health concerns prevalent among the elderly. Notably, Dementia related diseases, including Alzheimer's disease (AD) and frontotemporal dementia (FTD), significantly impair the quality of life for both affected seniors and their caregivers. However, the underlying neural mechanisms of these diseases remain incompletely understood, especially in terms of neural oscillations. In this study, we leveraged an open dataset containing 36 CE, 23 FTD, and 29 healthy controls (HC) to investigate these mechanisms. We accurately and clearly identified three stable oscillation targets (theta, ∼5 Hz, alpha, ∼10 Hz, and beta, ∼18 Hz) that facilitate differentiation between AD, FTD, and HC both statistically and through classification using machine learning algorithms. Overall, the differences between AD and HC were the most pronounced, with FTD exhibiting intermediate characteristics. The differences in the theta and alpha bands showed a global pattern, whereas the differences in the beta band were localized to the central-temporal region. Moreover, our analysis revealed that the relative theta power was significantly and negatively correlated with the Mini Mental State Examination (MMSE) scores, while the relative alpha and beta power showed a significant positive correlation. This study is the first to pinpoint multiple robust and effective neural oscillation targets to distinguish AD, offering a simple and convenient method that holds promise for future applications in the early screening of large-scale dementia-related diseases.

Details DOI

YNIMG Journal 2024 Journal Article

The neural oscillatory mechanism underlying human brain fingerprint recognition using a portable EEG acquisition device

Yuchen Lin
Shaojia Huang
Jidong Mao
Meijia Li
Naem Haihambo
Fang Wang
Yuping Liang
Wufang Chen

In recent years, brainprint recognition has emerged as a novel method of personal identity verification. Although studies have demonstrated the feasibility of this technology, some limitations hinder its further development into the society, such as insufficient efficiency (extended wear time for multi-channel EEG cap), complex experimental paradigms (more time in learning and completing experiments), and unclear neurobiological characteristics (lack of intuitive biomarkers and an inability to eliminate the impact of noise on individual differences). Overall, these limitations are due to the incomplete understanding of the underlying neural mechanisms. Therefore, this study aims to investigate the neural mechanisms behind brainwave recognition and simplify the operation process. We recorded prefrontal resting-state EEG data from 40 participants, which is followed up over nine months using a single-channel portable brainwave device. We found that portable devices can effectively and stably capture the characteristics of different subjects in the alpha band (8-13Hz) over long periods, as well as capturing their individual differences (no alpha peak, 1 alpha peak, or 2 alpha peaks). Through correlation analysis, alpha-band activity can reveal the uniqueness of the subjects compared to others within one minute. We further used a descriptive model to dissect the oscillatory and non-oscillatory components in the alpha band, demonstrating the different contributions of fine oscillatory features to individual differences (especially amplitude and bandwidth). Our study validated the feasibility of portable brainwave devices in brainwave recognition and the underlying neural oscillation mechanisms. The fine characteristics of various alpha oscillations will contribute to the accuracy of brainwave recognition, providing new insights for the development of future brainwave recognition technology.

Details DOI

YNIMG Journal 2021 Journal Article

Where does fear originate in the brain? A coordinate-based meta-analysis of explicit and implicit fear processing

Di Tao
Zonglin He
Yuchen Lin
Chang Liu
Qian Tao

Processing of fear is of crucial importance for human survival and it can generally occur at explicit and implicit conditions. It is worth noting that explicit and implicit fear processing produces different behavioral and neurophysiological outcomes. The present study capitalizes on the Activation Likelihood Estimation (ALE) method of meta-analysis to identify: (a) the "core" network of fear processing in healthy individuals; (b) common and specific neural activations associated with explicit and implicit processing of fear. Following PRISMA guidelines, a total of 92 fMRI and PET studies were included in the meta-analysis. The overall analysis show that the core fear network comprises the amygdala, pulvinar, and fronto-occipital regions. Both implicit and explicit fear processing activated amygdala, declive, fusiform gyrus, and middle frontal gyrus, suggesting that these two types of fear processing share a common neural substrate. Explicit fear processing elicited more activations at the pulvinar and parahippocampal gyrus, suggesting visual attention/orientation and contextual association play important roles during explicit fear processing. In contrast, implicit fear processing elicited more activations at the cerebellum-amygdala-cortical pathway, indicating an 'alarm' system underlying implicit fear processing. These findings have shed light on the neural mechanism underlying fear processing at different levels of awareness.

Details DOI